[Feature] Support async scheduling + PP (#32359)

Signed-off-by: yewentao256 <zhyanwentao@126.com>

vllm/v1/executor/multiproc_executor.py

@@ -411,9 +411,9 @@ class MultiprocExecutor(Executor):
 
     @cached_property
     def max_concurrent_batches(self) -> int:
-        if self.scheduler_config.async_scheduling:
-            return 2
-        return self.parallel_config.pipeline_parallel_size
+        # PP requires PP-size concurrent batches to fill the pipeline.
+        pp_size = self.parallel_config.pipeline_parallel_size
+        return 2 if pp_size <= 1 and self.scheduler_config.async_scheduling else pp_size
 
     def _get_output_rank(self) -> int:
         # Only returns ModelRunnerOutput from TP rank=0 and PP rank=-1